OpenAI previews privacy-focused system for detecting AI misuse
OpenAI previews Private Safety Processing, detecting cross-interaction AI misuse without exposing customer prompt content, rolling out in September.
OpenAI previewed Private Safety Processing for API and enterprise customers, a system that analyzes patterns across related interactions to flag potential misuse while restricting OpenAI personnel from accessing underlying content. It builds on Zero Data Retention (ZDR) deployments, keeping content on customer-controlled infrastructure or a planned option using customer-controlled encryption keys, with prompts and responses not retained after processing. A technical white paper and rollout are planned for September; images flagged as potential CSAM are the retention exception.
40